high performance computing on graphics processing units: hgpu.org

In this paper, we apply Particle-based Volume Rendering (PBVR) technique using a current programmable GPU architecture. Recently, the increasing programmability of GPU offers an efficient method of SIMD parallel algorithm to solve the speed problem. Due to the each point or pixel can be calculated independently, we use programmable graphics hardware to delegate all expensive rendering tasks to the GPU. Here we apply on the popular programming architecture CUDA based on GeForce 8800 graphics unit. This approach allows enormous volume particles to be rendered in SIMD way instead of time-costing sequence processing so that the rendering speed can be accelerated. In this processing, each particle can be handled separately by one of the multi-processors in GPU. In this implementation, we introduce a non-confliction way to map the conventional algorithm onto CUDA calculation architecture efficiently. We apply CUDA in the programming framework as a general purpose GPU calculation for PBVR instead of using conventional GPU pipeline. All the rendering flow can be divided into three stages: the beforehand data arrangement, particle projection and sub-pixel processing. In order to evaluate the performance, we compare the frame rate of GPU accelerated PBVR with traditional CPU based approach.